When a service works for the operator but not for a customer, repeating the same check from one location is often insufficient.
On August 7, the APNIC Blog published an analysis of partial reachability: a network or service remains accessible from some parts of the Internet but not from others. The authors describe these states as “peninsulas,” where a destination is reachable from one vantage point but not another.
In RIPE’s DNSMON data, effects linked to partial reachability could be five to almost ten times larger than events normally treated as meaningful outages. A small group of long-lived events—roughly 7% of all peninsula events—accounted for about 90% of the total time spent in peninsula states.
Possible causes include asymmetric filtering, peering or de-peering changes, and differences in routing policy. Reliable diagnosis needs several carefully chosen vantage points across different networks: a green check from one data center does not prove reachability for every customer.
This is a guest research post based on specific measurement datasets, not a universal Internet-wide statistic. The study covers IPv4; the authors leave IPv6 for future work. Partial reachability can have causes beyond BGP, so the symptom alone does not establish a root cause.